Industrial systems can create data from machinery, transportation vehicles, sensors, tools and interconnected assets. The challenge in this arena is less often data gathering. It is more complex to extract value
A reading, vibration, location or even a machine-event can be valuable, but with contextual analysis can reveal more information about the environment or conditions than on its own.
Artificial Intelligence of Things (AIoT) is a concept which pairs IoT data gathering with AI and analytics as enablers to extract value from systems and processes.
An Elementary Consideration on AIoT Architectures
An easy way to describe such architectures is as follows:
Connect → Collect → Analyze → Understand → Act
In each of these cases specific functions need to take place.
- Connecting sensors, machines or vehicles allows the data to be extracted or ingested.
There will be various methods depending on the environment. The essential point is enabling physical data to be accessed as electronic data.
- The next activity is data collection. This is about bringing the various data elements into the processing arena. Examples of these data elements might include:
• Temperature
• Vibration
• Pressure
• Location
• Movement
• Equipment Status
• Production Events
• Environmental Conditions
The issue with data collection is less about the presence of data (as shown by the wide range of examples) and more about consistency. Information coming from different sources might need contextual analysis.
- The value extraction process begins during analysis. This is not simply about extracting meaning from information (which is not always straightforward).
Analytics and AI can detect patterns or anomalies. Trends over time can point towards specific problems or even be used for diagnostics. For example, it might not be the immediate reading which points to an issue, but a shift in values or the comparison with other machines.
- Information which has been analyzed can then be enriched with additional contextual information.
A condition or event can be placed within an operational framework. The same machinery reading might be interpreted differently depending on location, additional maintenance or other factors. Information from several sources (or one) can support the creation of insights. Contextually enriched data is more likely to be accurate.
- With understanding and information, the final stage is to decide how to act. An action could be as simple as raising an alert or as complex as introducing a change to production processes. AIoT is not necessarily about removing humans from decision making but providing the right amount of information at the right time to benefit human activity.
A Practical Industrial Example of AIoT
Imagine a factory production space where numerous machines support the manufacturing process. Each has a set of sensors which report on vibration and temperature, but the plant has separate sets for maintenance schedules and another for production analytics.
By themselves, these systems do not communicate with each other, with each separate team seeing only the data relevant to their needs
An AIoT architecture (possibly paired with dedicated analytics) can enrich all available sources and identify patterns.
A gradual increase in vibration might not raise concerns by itself, but during analysis could be seen as problematic during a certain production period.
The issue is less about AI and more the introduction of interconnections.
It is also vital to remember that the AIoT environment does not replace operators (and decision making). As mentioned previously, value comes from adding to information. In some industrial situations it is about raising concerns or providing the appropriate data to take other action.
Opportunities Across Several Industrial Sectors
The same principle can apply to different environments. The general function for industrial spaces is as follows:
Making data gathered from several sources (machines, processes, transport vehicles, equipment and tools) support operational decisions.
Such decisions can be seen across several sectors and areas:
• Manufacturing – equipment, production, process and related aspects.
• Logistics – connected vehicles, assets, tracking and location.
• Construction – equipment, materials, workers and projects.
• Energy and Utilities – asset management, equipment or operational analytics.
• Mining – operations, asset usage, environmental or production data.
The exact functions will be different for each case, but the overall challenge remains the same: extract value from seemingly inaccessible data sources.
The Importance of Addressing a Concrete Problem
One common failure of digital programs in industrial settings is to begin with technology, not an operation. Adding more sensors does not necessarily create new value, nor does an AI model analyzing unstructured data-points gather meaningful results
A more efficient method is to identify a specific challenge beforehand and building suitable analytics around it.
A practical consideration is: What operational decision am I trying to support?
From there it is about analyzing what is necessary. An effective AIoT venture is more about addressing challenges than simply adopting new solutions.
This is also the approach of Aperture Venture Studio which focuses on the creation of AIoT and Physical AI ventures which take into account day-to-day processes and needs: Aperture Venture Studio.
Conclusion: Building Meaningful Interconnections
At its very essence AIoT is less about connecting vast technological universes and more about how interconnections generate meaningful data relations.
Sensors open environments, connectivity moves this data and analytics identify patterns, value and support decisions. Contextual information adds to this process while actual people or automated workflows perform actions.
For organizations which want to begin the journey of adopting industrial AIoT solutions an effective way forward is to analyze one concrete opportunity. How can data help to improve a specific decision? What sources of information need to be interconnected? What kind of analyses support higher levels of understanding? At this point, connected data can move towards generating industrial intelligence.
Top comments (0)